基于毛竹叶片理化参数的刚竹毒蛾危害检测研究

虫害检测算法研究是开展虫害快速、准确监测，制定精准森防检疫措施的重要基础。以毛竹叶片为研究尺度，基于刚竹毒蛾危害下的寄主外部形态与内部生理现象总结，选择并实测叶损量LL、相对叶绿素含量RCC、相对含水量RWC、原始光谱的733.66~898.56 nm值(ρ_{733.66~898.56})、一阶微分光谱的562.95~585.25 nm值(ρ′_{562.95~585.25})与706.18~725.41 nm值(ρ′_{706.18~725.41})等理化参数，随机划分实验组(63组)和验证组(37组)并设计5次重复实验；分别运用Fisher判别分析、BP神经网络、随机森林等三种方法建立刚竹毒蛾危害等级的检测模型，从检测精度、Kappa系数及R2等指标对模型的检测效果予以分析和比较。结果显示，Fisher判别分析、BP神经网络、随机森林的检测精度分别为69.19%，65.41%，83.78%，Kappa系数分别为0.576 9，0.532 4和0.778 8，R2分别为0.722 2，0.582 6和0.870 9，总体而言，三种方法均具备刚竹毒蛾危害的检测能力，随机森林的检测效果最优，Fisher判别分析次之，再次为BP神经网络；从分等级来看，随机森林的检测精度亦优于Fisher判别分析与BP神经网络，但3种方法对中度危害等级的检测精度均有所不足。该成果可为刚竹毒蛾危害及其他病虫害检测算法的选择提供参考，并为进一步建立冠层、遥感影像像元等尺度的虫害检测模型奠定基础。

Pantana Phyllostachysae Chao Damage Detection Based on Physical and Chemical Parameters of Moso Bamboo Leaves

Pest detection algorithm research is an important guarantee to precisely and rapidly monitor the forest pest and forest protection and quarantine. Based on the external morphology of the host and its internal physiological phenomena, taking the leaf loss (LL), relative chlorophyll content (RCC), relative water content (RWC), and the three spectral values of the characteristic wavelengths (ρ_{733.66～898.56}, ρ′_{562.95～585.25}, ρ′_{706.18～725.41}) as the experimental data which were randomly divided into experimental group (63) and verificantion group (37) with 5 repeated tests, then the models of Fisher discriminant analysis, random forest and BP neural networks for pest levels were constructed. The detection accuracy, Kappa coefficient and R2 were used to comprehensively compare the detection effects of these three algorithms. The results showed that the detection accuracy of Fisher discriminant analysis, BP neural networks and random forest were 69.19%, 65.41% and 83.78%, and Kappa coefficient were 0.576 9, 0.532 4 and 0.778 8, and R2 were 0.722 2, 0.582 6 and 0.870 9. Overall, all of these algorithms have the capability of pest detection, among which, the detection effect of the random forest is the best, and Fisher discriminant analysis is secondly, and BP neural networks is thirdly. Besides, the accuracy of random forest detection is superior to that of Fisher discriminant analysis and BP neural networks in non-damage, mild damage and severe damage, but these three methods have insufficient detection accuracy for moderate damage level. The results could be a reference tothe selection of detection algorithm in P. chao and other types of diseases and insect pests, building a strong foundation for further study.